data
SequenceClassificationOutput
extends
ModelOutputSequenceClassificationOutput(logits: Tensor, loss: Tensor | None = None, hidden_states: tuple[Tensor, ...] | None = None, attentions: tuple[Tensor, ...] | None = None)Output of any whole-sequence classification head.
Attributes
logitsTensorPer-sequence class logits, shape
(B, num_labels). A
num_labels of 1 marks a regression head, whose loss is MSE
rather than cross-entropy.loss(Tensor or None, optional)Scalar loss when
labels were supplied.hidden_states(tuple[Tensor, ...] or None, optional)Per-layer hidden states.
attentions(tuple[Tensor, ...] or None, optional)Per-layer attention weights.
Notes
Returned by every {Family}ForSequenceClassification head. It
exists so a caller can tell a (B, num_labels) sentence logit
apart from a (B, T, vocab_size) masked-LM logit by type rather
than by inspecting shapes.
Examples
>>> import lucid
>>> from lucid.models import SequenceClassificationOutput
>>> out = SequenceClassificationOutput(logits=lucid.zeros(1, 2))
>>> out.logits.shape # (B, num_labels)
(1, 2)